Matrix Heterogeneity and Host–Parasitoid

Ecology, 84(6), 2003, pp. 1506±1516 ᭧ 2003 by the Ecological Society of America

MATRIX HETEROGENEITY AND HOST±PARASITOID INTERACTIONS IN SPACE

JAMES T. C RONIN1 Department of Biological Sciences, 206 Life Sciences Building, Louisiana State University, Baton Rouge, Louisiana 70803-1715 USA

Abstract. In this study, I experimentally examined how the landscape matrix in¯uenced the movement, oviposition behavior, and spatial distribution of Anagrus columbi, a common egg parasitoid of the planthopper Prokelisia crocea. Both species exist among discrete patches of prairie cordgrass (Spartina pectinata), the sole host plant of P. crocea. Based on out-planted cordgrass pots bearing host eggs (to assess parasitism) or sticky leaves (traps for adult A. columbi), I found that the distribution of adult female A. columbi and pattern of ovipositions within a cordgrass patch were strongly matrix dependent. Female densities were 59% lower on the edge than interior of patches embedded in a mud¯at matrix, but were evenly distributed within patches embedded in a matrix consisting of either native grasses or the exotic grass smooth brome (Bromus inermis). In contrast, parasitism was higher in the interior than edge for patches in all three matrix types. The lack of corre- spondence between A. columbi density and parasitism was attributed to differences in oviposition behavior: A. columbi parasitized 71% more hosts per capita in the interior than edge for patches embedded in nonhost grasses, but equal numbers on the edge and interior of patches embedded in mud¯at. Matrix-dependent differences in the within-patch distribution and oviposition behavior of A. columbi can in¯uence the distribution of parasitism risk and host±parasitoid stability at the patch level. Matrix composition also affected the pattern of movement through the matrix and the colonization of nearby cordgrass patches. Anagrus columbi females emigrating from a mud¯at-embedded patch were captured at very low, but constant, numbers with distance out into the matrix, suggesting that they were reluctant to enter or remain in the mud¯at. In contrast, A. columbi females entering a nonhost grass matrix had numbers that were high near the patch border and then declined exponentially with distance. These patterns of movement were likely responsible for the very different colonization rates for experimental patches embedded in different matrix types and located 3 m from a source patch of A. columbi. Patches embedded in brome were colonized at a rate that was 3.0 and 5.7 times higher than for patches in native grass or mud¯at, respectively. Finally, based on a census of cordgrass patches spanning ®ve generations, A. columbi densities and proportion of patches occupied generally increased with increasing host density, patch isolation, and the proportion of the surrounding matrix that was mud¯at. Patch size had no effect on the distribution of A. columbi. Overall, these data suggest that cordgrass patches in a nonhost grass matrix, particularly smooth brome, have high connectivity relative to patches in a mud¯at matrix. Changes in connectivity due to changes in matrix composition can signif- icantly in¯uence host±parasitoid persistence at the metapopulation level. Key words: Anagrus; connectivity; dispersal; edge effect; egg parasitoid; landscape matrix; metapopulation; patch dynamics; patch occupancy; planthoppers; Spartina; tallgrass prairie.

INTRODUCTION lation, quality, or density of predators or prey (Hanski 1999). This traditional approach ignores the possibility Among the numerous theoretical studies of predator± that the matrix within which these patches are embedded prey dynamics, the consensus is that spatial heterogeneity may in¯uence dispersal processes and hence, the con- can signi®cantly affect the temporal dynamics, stability, nectivity among patches (Taylor et al. 1993, Roland et and spatial distribution of populations (reviewed in Til- al. 2000, Ricketts 2001). From the vantage point of land- man and Kareiva 1997, Hanski 1999, Hassell 2000). The scape ecology, treatment of the matrix as homogeneous movement of individuals among patches, which is critical to these models, is usually assumed to be affected only and unimportant is arguably a major ¯aw in metapopu- by the properties of the patch itself; i.e., patch size, isolation theory (Wiens et al. 1993, Wiens 1997; but see Molainen and Hanski 1998). For insect herbivores distributed among host-plant Manuscript received 26 June 2002; revised 2 October 2002; patches, the matrix is often heterogeneous, consisting accepted 16 October 2002. Corresponding Editor (ad hoc): J. A. Rosenheim. of a complex mosaic of land cover types. Some matrix 1 E-mail: [email protected] types may be resistant and favor low rates of interpatch 1506 June 2003 PARASITOID RESPONSE TO THE MATRIX 1507 movement whereas others may be benign (i.e., have (sometimes inhabited by saltwort [Salicornia rubra low resistance) and favor high rates of movement. For Nels.], (2) a mixture of native grass species (primarily a single species, the level of matrix resistance may Andropogon scoparius Michx., A. gerardii Vitman, and dictate whether the ensemble of subpopulations form Agropyron smithii Rydb.), and (3) the exotic grass, a metapopulation or a single patchy population (Har- smooth brome (Bromus inermis Leyss). Brome is sim- rison and Taylor 1997). For a predator and its prey, ilar in stature and appearance to cordgrass, and both differential effects of the matrix on their respective species are taller than most native grasses (Wilson and dispersal abilities can profoundly impact regional sta- Belcher 1989). Planthopper adults are 90% macrop- bility (e.g., Reeve 1988, Comins et al. 1992). Reeve terous and quite mobile. The number of immigrants (1988), for example, demonstrated that host±parasitoid increases with patch area (even single-stem patches are interaction persistence can be maintained in the ab- colonized and oviposited within) and is generally un- sence of regulatory processes if parasitoid dispersal affected by patch isolation (Cronin 2003b). In a series among patches is greater than that of its host. Whereas of ®eld experiments, K. J. Haynes and J. T. Cronin a number of studies have identi®ed the effects of matrix (unpublished manuscript) found that planthopper em- composition on the movement patterns of herbivores igration and immigration rates were highest for patches among host-plant patches (e.g., Kareiva 1985, Jonsen embedded in nonhost grasses and lowest for patches et al. 2000, Roland et al. 2000), no studies have ex- in mud¯at. We also found a matrix-dependent edge perimentally quanti®ed its effects on interacting spe- effect: planthoppers were generally more abundant at cies. the patch edge relative to the interior for patches in In addition to in¯uencing dispersal, matrix compo- mud¯at, but not for patches in nonhost grasses. sition is likely to also affect the among- and within- In this study, I experimentally examined how the patch distributions of herbivores and their enemies. Oc- landscape matrix in¯uences the movement and spatial cupancy rates and densities are likely to be higher for distribution of Anagrus columbi Perkins (Hymenop- patches embedded within a low- than a high-resistance tera: Mymaridae), a common egg parasitoid of P. cro- matrix (e.g., Kareiva 1985, Jonsen et al. 2000). In the cea (Cronin 2003a). First, I assessed whether an edge former matrix type, high immigration should promote effect (in terms of A. columbi density, per capita ovi- the rescue effect and the rapid recolonization of vacant position rate, and proportion of hosts parasitized) oc- patches (Brown and Kodric-Brown 1977, Hanski curred among cordgrass patches and whether those ef- 1999). Within a patch, some matrix types may inhibit fects were matrix dependent. Second, I estimated A. emigration (i.e., the patch edge is impermeable or re- columbi colonization rates of experimental cordgrass ¯ecting; Stamps et al. 1987) and cause organisms to patches that were positioned 3 m away from natural accrue near the patch boundary (Cantrell and Cosner cordgrass patches embedded in each of the three matrix 1999). Other matrix types may foster high patch per- types. Third, patterns of spatial spread through each meability and the absence of a density edge effect. The matrix type were quanti®ed. Fourth, I determined the study of edge effects has grown in recent years with effects of landscape structure (patch size, isolation, and the recognition that these effects can signi®cantly in- matrix composition) on A. columbi density and occu- ¯uence movement patterns, species interactions, and pancy among patches using ®ve generations of census community structure (Fagan et al. 1999). Edge effects data. Finally, I addressed how the response to the ma- have been commonly reported for insect herbivores trix by A. columbi and its host might impact patch (e.g., Cappuccino and Root 1992, Roland and Kaupp connectivity and the spatial and temporal population 1995, Cappuccino and Martin 1997), and in some cases dynamics of this host±parasitoid interaction. there is evidence that the effects are transmitted to higher trophic levels (e.g., Lovejoy et al. 1984, Bolger METHODS et al. 2000). If the edge effect is moderated by the type Life history of matrix, predator±prey interactions may be landscape dependent (see Tscharntke et al. 2002). To advance our Prokelisia crocea is a monophagous phloem-feeder understanding of predator±prey interactions at the land- and dominant herbivore of prairie cordgrass throughout scape level, studies are needed that explore how the the plant species' range (Holder and Wilson 1992, Cro- matrix in¯uences aspects of interpatch movement (e.g., nin 2003a, b). In North Dakota, planthoppers overwin- emigration, immigration, patterns of spatial spread) and ter as ®rst instar nymphs, reach peak adult densities in spatial distributions for both species. early June, and then lay eggs beneath the adaxial sur- In the tall-grass prairies of the midwestern United face of cordgrass leaves (Cronin 2003a, b). A second States, the planthopper Prokelisia crocea (Van Duzee) generation follows, with adults peaking in early Au- (Hemiptera: Delphacidae) is distributed among very gust. The most obvious natural enemy of P. crocea is discrete patches of its host plant, prairie cordgrass A. columbi. Parasitism rates per cordgrass patch range (Spartina pectinata Link [Poaceae]) (Cronin 2003a, b). from 0% to 100%, with an average of 21% (Cronin Within the prairie landscape, cordgrass patches are em- 2003a). In general, A. columbi searches for hosts at bedded in one of three basic matrix types: (1) mud¯ats random within a cordgrass leaf, demonstrates no ability 1508 JAMES T. CRONIN Ecology, Vol. 84, No. 6

7 d to capture A. columbi and accumulate parasitism. After the pots were collected and returned to the laboratory, I censused the 30 cordgrass patches for stem densities, host densities, and parasitism. I placed a 0.25 ϫ 0.25 m sampling frame in three randomly cho- sen locations along the interior and edge transects, counted the number of cordgrass stems, and collected all of the leaves that were naturally infested with planthopper eggs. Stem densities (on a per square meter basis) at each transect location within a patch were determined from the average number of stems per sampling frame. In the laboratory, the sticky leaves from trap pots were examined with a dissecting microscope and the number of male and female A. columbi counted. After a 5-d incubation period, the leaves from host pots and naturally infested leaves from the census were dis- sected to determine the number of healthy and parasitized hosts. I note here that at the time of this study, planthoppers were almost entirely in the egg stage. Therefore, eggs laid by resident planthoppers onto potted cordgrass plants did not take place with any mea- FIG. 1. Diagram of matrix experiment. Potted cordgrass with planthopper eggs (open circles) or Tanglefoot (closed surable frequency. circles) on leaves formed each transect (interior, edge, and For each transect location, I used the data obtained matrix). The design was replicated 10 times in each of three from the experimental host pots and trap pots to com- matrix types: mud¯at, native grasses, or brome. pute the following transect means (on a per-leaf basis): (1) number of female A. columbi immigrants, (2) pro- to avoid superparasitism, and displays a type I func- portion of hosts parasitized, and (3) number of hosts tional response over the range of hosts normally found parasitized per female A. columbi (mean number of in the ®eld (Cronin 2003a). Although A. columbi has hosts parasitized per leaf/mean number of A. columbi been recorded as a parasitoid of several planthopper captured per leaf). Finally, for each transect separate species (Krombein et al. 1979), in northeastern North measures of mean host egg density, number of hosts Dakota, it appears to attack only P. crocea residing parasitized, and proportion parasitized were deter- within cordgrass patches (Cronin 2003a). mined for the naturally occurring planthopper-infested leaves. Matrix experiment The effect of matrix type within which a patch was This experiment was designed to quantify the dis- embedded and transect location (interior, edge, or ex- tribution of A. columbi adults and parasitized hosts terior) on A. columbi density, proportion parasitized, within cordgrass patches embedded in different matrix and per capita host parasitized were evaluated with types, and to estimate the effects of different matrix separate pro®le ANOVAs. Here, pro®le ANOVA is a types on colonization rates of experimental cordgrass multivariate test (comparable to a repeated-measures patches. I selected 10 large cordgrass patches (at least ANOVA; Simms and Burdick 1988) that allows for the 40 m2 in area) that were embedded in each of three three transects associated with each patch to be non- matrix types (mud¯at, native grass, brome). All patches independent (they share a common environment and had a very distinct cordgrass-matrix edge and were are potentially correlated in A. columbi density and separated from each other by a minimum of 50 m. parasitism). The responses within each transect (e.g., Given the limited dispersal ability of A. columbi (Cro- proportion of hosts parasitized) were the dependent nin 2003a), the use of large patches was expected to variables and the matrix type was the independent var- lessen the importance of patch shape in affecting this iable. Density and per capita parasitized were ln-trans- parasitoid's movement and distribution within a patch. formed and proportion parasitized was arcsine square- Transects of potted cordgrass plants were placed in root transformed to achieve normality and homogeneity parallel at 2 m into the interior, the edge, and3mout of variances. To evaluate differences between two ma- into the matrix of each patch (Fig. 1). Within a transect, trix types (e.g., brome vs. mud¯at), I used the same host pots (pots with host-infested leaves; n ϭ 3) and pro®le ANOVA model but with only these two matrix trap pots (pots with sticky leaves; n ϭ 3) were used to types included. Contrasts between the edge and interior assess A. columbi parasitism and density per leaf, re- (edge effects) and the interior and matrix were eval- spectively. The creation of these pots and their validity uated for patches embedded in each matrix type using in assessing these parasitoid traits are described in Ap- paired t tests. To ensure that the Type I error rate for pendix A. Host and trap pots were left in the ®eld for each pro®le ANOVA and contrast (within a matrix) did June 2003 PARASITOID RESPONSE TO THE MATRIX 1509

not exceed the nominal rate of 0.05, Bonferroni se- Finally, I compared the distribution of Pjd for each quential corrections were used to assess signi®cance matrix type with the pattern of spatial spread predicted (Sokal and Rohlf 1995). by a simple diffusion model. Turchin and Thoeny Finally, the ability of A. columbi to successfully dis- (1993) provided the derivation for the following ana- perse from one of the large natural cordgrass patches lytical formula for predicting spatial spread based on

(10 per matrix type) to an experimental patch located cumulative recaptures (see also Okubo 1980): Pjd ϭ 3 m into the matrix was obtained from trap-plant cap- AdϪ1/2eϪd/B. Here, A is a scaling parameter and is pro- tures. Because the 10 cordgrass patches per matrix were portional to the product of the source population and relatively isolated from other native cordgrass (mini- the recapture ef®ciency. B is a measure of the spatial mum of 50 m), each likely represented the primary scale of dispersal and is equal to the square root of the source of A. columbi that colonized the potted cord- ratio of the diffusion rate and the disappearance rate grass associated with it. Therefore, my index of col- (death or emigration from the experimental area). An onization rate was the ratio of mean A. columbi captures insect population with a large value of B would have per leaf in the matrix transect and the mean captures a greater dispersal range than one with a smaller B. per leaf within the source patch ([interior mean ϩ edge The model above has the linear form, ln(Pjd) ϩ ½ln(d) mean]/2). Because the distribution of this index could ϭ ln(A) Ϫ d/B, and can be ®t using least-squares re- not be normalized (due to a disproportionate number gression (Sokal and Rohlf 1995). of zeros and a right skew), the effect of matrix composition on the colonization rate was evaluated with a Landscape effects on spatial distribution Kruskal-Wallis test (Sokal and Rohlf 1995). Mann- The effects of landscape structure (patch size, patch Whitney U tests (Bonferroni-corrected) were used for isolation, and matrix type) and planthopper egg density all pairwise contrasts. on the distribution of A. columbi among cordgrass patches was evaluated from census data that spanned Movement through the matrix ®ve generations (mid-1999 through the end of 2001) and 147 cordgrass patches. The census took place in a To assess how the matrix in¯uences interpatch move- single prairie fragment (Site 104), a 65-ha area adjacent ment of A. columbi, I performed an experiment that to Kelly's Slough National Wildlife Refuge in north- quanti®ed the pattern of spatial spread from cordgrass eastern North Dakota (47.94184 N, 97.31036 W). Ev- patches embedded in a mud¯at or native grass matrix. ery generation, I ascertained the presence/absence and Large cordgrass patches were selected from the edge density of P. crocea (eggs) and A. columbi (parasitized of a prairie fragment. Yellow sticky traps (8 ϫ 13 cm hosts) for each cordgrass patch (see Appendix B for Dayglo Saturn yellow index cards coated with Tangle- details). In addition, each patch was characterized by foot, Tanglefoot Company, Grand Rapids, Michigan) its size (measured in square meters), isolation from attached to the tops of 0.5-m tall PVC poles were used nearest neighbor patches, and composition of the sur- to capture dispersing A. columbi. The traps were po- rounding matrix (Appendix B). Because mud¯at was sitioned along a transect at 5 m within the patch, the the most different matrix with regard to A. columbi patch edge, and at 2, 5, 10, 20, and 30 m away from distributions (relative to the two nonhost grass matri- the patch (and in the opposite direction of other native ces; see Results), I chose to use the proportion of buffer cordgrass patches). Trap transects were placed in as- habitat composed of mud¯at as my index of matrix sociation with six mud¯at- and six native grass-bor- composition (see also Moilanen and Hanski 1998). For dered patches and were deployed in the ®eld for 2 wk, each generation, I quanti®ed the effect of landscape corresponding to the peak activity of A. columbi. The structure (patch size, patch isolation, and proportion of number of female A. columbi captured in each transect matrix that is mud¯at) and planthopper egg density on j and trap distance d, Cjd, was converted to a proportion: two dependent variables: A. columbi density and patch Pjd ϭ Cjd/Nj; where Nj ϭ total number of A. columbi occupancy. Least-squares regression was used for the captured in transect j. former, and logistic regression for the latter, dependent The effect of matrix type (mud¯at or native grass) variable (details are provided in Appendix B). and trap distance on the proportional recaptures (ln

[Pjd]) was evaluated with a pro®le ANOVA. As in the RESULTS previous experiment, traps located at different distances from the same patch could not be considered sta- Matrix experiment tistically independent. Tests for differences between On average, A. columbi density (number of females matrix types at each trap distance (e.g., mud¯at edge per trap-pot leaf) within a patch was independent of vs. native grass edge) were evaluated with separate the composition of the surrounding matrix (Table 1, two-sample t tests (Sokal and Rohlf 1995). The critical Fig. 2A). However, the distribution of A. columbi with- value of ␣ for each test was adjusted using a sequential in a patch was very strongly matrix dependent, as in- Bonferroni correction to achieve an overall error rate dicated by the signi®cant matrix ϫ transect interaction of 0.05. (Table 1). For patches embedded in a mud¯at, densities 1510 JAMES T. CRONIN Ecology, Vol. 84, No. 6

TABLE 1. Results from separate pro®le ANOVAs for the icantly different (Mann-Whitney U statistic ϭ 53.00, effect of matrix type and transect location within a patch on ln(parasitoid density), angular-transformed proportion df ϭ 1, P ϭ 0.009). of hosts parasitized, and ln(per capita parasitized). In contrast to the density distribution of A. columbi, the distribution of parasitism was matrix dependent Source of variation df MS FP(Table 1, Fig. 2B). The proportion of hosts parasitized Parasitoid density per leaf averaged (all transects combined) 0.17 Ϯ 0.02, Among patches 0.12 Ϯ 0.03, and 0.06 Ϯ 0.02 for patches in mud¯at, Matrix 2 14.43 3.57 0.042 native grass, and brome, respectively. Only the pro- Error 27 4.04 3.33 portion for patches in mud¯at and brome was signi®- Within patches cantly different (F1,18 ϭ 7.42, P ϭ 0.014). For all three Transect 2 19.80 13.22 Ͻ0.001² matrix types, parasitism declined signi®cantly from the Transect ϫ matrix 4 8.01 5.35 0.001² Error 54 1.50 Proportion parasitized Among patches Matrix 2 0.09 6.63 0.005² Error 27 0.01 Within patches Transect 2 0.32 33.21 Ͻ0.001² Transect ϫ matrix 4 0.04 4.63 0.003² Error 54 0.01 Per capita parasitized Among patches Matrix 2 6.21 1.63 0.273 Error 6 3.82 Within patches Transect 2 23.24 21.46 Ͻ0.001² Transect ϫ matrix 4 8.09 7.47 0.003² Error 12 1.08 Notes: A Type I error rate for all three ANOVAs combined was maintained at ␣ϭ0.05 by using a sequential Bonferroni correction. A signi®cant P value is indicated with ².

declined by 59% from the interior to the edge (based on the mean proportional difference between transect pairs), a signi®cant density edge effect (Fig. 2A, Ap- pendix C). In the mud¯at matrix, A. columbi density was 84% lower than in the patch interior. In contrast, there was a gradual decline in A. columbi density between the interior and matrix for patches in native grasses (no edge effect evident; Appendix C), and no change in density among transects for the patches in brome (Fig. 2A). Overall, the density distribution of A. columbi adult females among transects within a patch was indistinguishable between patches embedded in mud¯at and native grass (as determined from the transect ϫ matrix interaction term from a pro®le AN- OVA that included only these two matrix types;

F2,36 ϭ 1.95, P ϭ 0.158) and between native grass and brome (F2,36 ϭ 3.17, P ϭ 0.054); but the difference was signi®cant between mud¯at and brome (F2,36 ϭ 11.51, P Ͻ 0.001). The colonization rate index for experimental cordgrass patches located 3 m away from a source patch was matrix dependent (Kruskal-Wallis statistic ϭ 6.82, FIG. 2. (A) Parasitoid densities, (B) proportion of hosts df ϭ 2, P ϭ 0.030). Patches in brome had the highest parasitized, and (C) per capita number of hosts parasitized for patches embedded in three different matrices (mud¯at, a rate (0.85 Ϯ 0.27; mean Ϯ 1 SE), followed by native mixture of native grass species, and smooth brome). Patch grasses (0.29 Ϯ 0.14), and ®nally mud¯at (0.10 Ϯ positions include 2 m into the interior, the edge, and 3 m into 0.05). Only the rate in brome and mud¯at was signif- the matrix. Means Ϯ 1 SE are reported. June 2003 PARASITOID RESPONSE TO THE MATRIX 1511

interior and edge, could have been in¯uenced by the density of hosts present on naturally occurring cordgrass plants. (Anagrus columbi larvae in naturally occurring hosts were too immature to represent a source of adult parasitoids during the weeklong study; thus, they likely did not play a part in causing the patterns observed in Fig. 2.) Egg densities on natural plants (numbers/stem) were independent of location within a patch, but highly dependent upon matrix type (Fig. 3). Despite these differences, pro®le ANOVAs (matrix type and transect within a patch as the main effects) performed with and without the density of naturally occurring hosts (ln [mean of edge and interior sam- ples]) included as a covariate led to the same conclu- sions (not reported). Therefore, naturally occurring hosts had no effect on the results reported above. Movement through the matrix FIG. 3. Natural host-egg densities (mean per stem Ϯ SE) in relation to the type of matrix within which a patch is The pattern of A. columbi captures-with-distance dif- embedded and transect location (2 m into the patch interior fered fundamentally between a native grass and mud¯at or at the patch edge). Based on a pro®le ANOVA, ln(egg matrix, based on the signi®cant matrix ϫ trap distance densities) were signi®cantly affected by the matrix (F 2,27 ϭ interaction term (Fig. 4). In the grass matrix, captures 4.41, P ϭ 0.022) but not by transect location (F1,27 ϭ 2.63, P ϭ 0.117). There was no interaction between the matrix and declined gradually from the patch interior to 30 m out transect location (F2,27 ϭ 0.98, P ϭ 0.388). into the matrix. There was no difference in A. columbi captures between the patch interior and edge of grass- bordered patches (Fig. 4), corroborating ®ndings in the patch interior to 3 m out into the matrix (Table 1, Fig. previous experiment. The linear form of the simple 2B). On average, parasitism was 48% higher on the diffusion model provided a signi®cant ®t to the combined patch interior than the edge. A signi®cant edge effect capture data in the grass matrix: for distances between 2 in parasitism was found for patches in mud¯at and and 30 m from the patch edge, y ϭϪ0.04x Ϫ 0.96 native grass but not for patches in brome (Appendix (R2 ϭ 0.32, P Ͻ 0.001). The addition of a quadratic C). As with A. columbi density, I found that the dis- term did not improve the model ®t. For mud¯at-em- tribution of parasitism among transects was matrix de- bedded patches, the proportion captured dropped steep- pendent (signi®cant matrix ϫ transect interaction; Ta- ble 1). Parasitism declined more steeply from the interior to the matrix for mud¯at-, than for native grass or brome-embedded patches; however, no signi®cant difference could be found between any two matrix types. Finally, the per capita hosts parasitized was similar across matrix types, but declined signi®cantly with distance from the patch interior to the matrix (Table 1, Fig. 2C). For all matrix types combined, an average of 44.1 Ϯ 13.0 hosts were parasitized per female A. columbi per leaf in the patch interior, 22.4 Ϯ 6.0 hosts parasitized at the edge, and 12.8 Ϯ 4.7 hosts parasitized in the matrix. Matrix type affected the per capita parasitized only through its interaction with transect location (Table 1). Most notably, a signi®cant edge effect in the per capita parasitized was found for patches em- FIG. 4. The proportion of A. columbi captured with re- bedded in native grass and brome, but not in mud¯at spect to distance from the patch edge for two matrices, mud- ¯at and native grass. Means Ϯ 1 SE are reported. Based on (Appendix C). Differences in the distribution of female a pro®le ANOVA, trap distance and the interaction between

A. columbi may have explained these patterns. Within trap distance and matrix type (F6,49 ϭ 68.90, P Ͻ 0.001 and a patch (all matrix types combined), the per capita hosts F6,49 ϭ 14.54, P Ͻ 0.001; respectively), but not matrix type parasitized was negatively correlated with female A. (F1,10 ϭ 4.34, P ϭ 0.064), affected the proportion captured (ln-transformed prior to analysis). Signi®cant differences be- columbi density (R ϭϪ0.41, n ϭ 60, P ϭ 0.020). tween matrix types at a given distance are denoted by *** Differences in the distribution of A. columbi females, (in all cases P Ͻ 0.001; based on separate two-sample t tests parasitism and per capita parasitized between the patch with a sequential Bonferroni correction of ␣). 1512 JAMES T. CRONIN Ecology, Vol. 84, No. 6

TABLE 2. The effect of landscape structure (patch size, isolation, and percentage adjacent matrix that is mud¯at) and host density (eggs/stem) on A. columbi density (parasitized hosts/ stem) or whether or not a patch was occupied by A. columbi.

Signi®cance of independent variables Model statistics Genera- Patch A. columbi tion n R2² P value size Isolation %Mud Hosts Density 1999, 2 25 0.593 Ͻ0.001 0.029 0.849 0.462 0.010 2001, 1 86 0.640 Ͻ0.001 0.456 0.015 0.010 Ͻ0.001 2000, 2 86 0.725 Ͻ0.001 0.233 0.003 Ͻ0.001 Ͻ0.001 2001, 1 82 0.900 Ͻ0.001 0.773 0.016 Ͻ0.001 Ͻ0.001 2001, 2 ´´´ ´´´ ´´´ ´´´ ´´´ ´´´ ´´´ Occupancy 1999, 2 ´´´ ´´´ ´´´ ´´´ ´´´ ´´´ ´´´ 2000, 1 84 0.102 0.134 0.528 0.171 0.955 0.097 2000, 2 89 0.430 Ͻ0.001 0.819 0.008 0.042 0.009 2001, 2 73 0.454 Ͻ0.001 0.150 0.282 0.008 0.002 2001, 2 104 0.111 0.049 0.469 0.913 0.965 0.019 Notes: Least-squares regression was used for the continuous variable parasitoid density, and logistic regression was used for patch occupancy frequencies. To control for Type I errors associated with conducting two tests per generation on the same data set, the critical value of ␣ for rejection of the null hypothesis was set at 0.025. P values in bold were deemed statistically signi®cant. ² For logistic regression analyses, McFadden's ␳2 is reported instead of R2. ly from the interior to 5 m out into the matrix, and then ence on the likelihood that a cordgrass patch was oc- remained relatively constant with distance out to 30 m. cupied by A. columbi (given that hosts were present) Because captures did not continue to decline with dis- was host density (Table 2). For all but one generation tance, the slope of the simple diffusion model was not (2000, generation 1, P ϭ 0.097), the frequency of signi®cantly different from zero (y ϭ 0.03x Ϫ 2.03; patches occupied by A. columbi increased signi®cantly R2 ϭ 0.11, P ϭ 0.087). with increasing host density. Overall, the logistic regression model explained an average of 27 Ϯ 10% of Landscape effects on spatial distribution the variation in presence/absence of A. columbi among The distribution of A. columbi among cordgrass cordgrass patches (based on McFadden's ␳2). patches was well ®t by a model that included features DISCUSSION of the landscape (patch size, patch isolation, matrix composition) and host egg density. Among the four Matrix effects on A. columbi movement generations that tests were possible, the model ex- and distribution plained an average of 71.5 Ϯ 6.8% of the variation in In the tall-grass prairies of the Great Plains, the land- A. columbi distributions among patches (based on mod- scape matrix appears to have a signi®cant effect on A. el R2; Table 2). Clearly the most important factor in columbi movement and spatial distribution within and the model was host egg density. In all generations, A. among cordgrass patches. Within mud¯at-bordered columbi densities increased signi®cantly with increas- patches, A. columbi densities were high in the patch ing numbers of hosts, and this variable alone explained interior relative to the patch edge, and were extremely 59.2 Ϯ 5.0% of the total variation in parasitoid den- low in the mud¯at matrix. In contrast, no edge effect sities. The proportion of the matrix that was bare in A. columbi density was evident in matrix composed ground (mud¯at) and patch isolation had signi®cant of nonhost grasses; and for brome-embedded patches, effects on A. columbi density (independent of host A. columbi densities were just as high 3 m out in the abundance), explaining 7.4 Ϯ 3.3% and 4.9 Ϯ 3.0% of matrix as in the patch interior. These data suggest that the total variation in density, respectively (Table 2). the mud¯at edge is not only hard, but possibly repellent On average, A. columbi densities increased with an (Stamps et al. 1987). Brome edges, and to a lesser increase in the proportion of the matrix that was mud¯at extent native grass edges, appear softer and more per- and with isolation. Patch size had no effect on A. col- meable. Brome is similar in stature and appearance to umbi numbers per cordgrass stem. cordgrass and it is possible that A. columbi does not Patch occupancy frequencies changed substantially readily distinguish between these two plant species. over the course of this study. Of the host-infested patch- The reason that brome has a low resistance to emi- es, 100% was occupied by A. columbi in 1999 (con- grating A. columbi remains unknown. sequently no logistic regression was performed), 86% For patches embedded in all matrix types, parasitism in 2000 (generation 1), 88% in 2000 (generation 2), was generally greater on the patch interior than the 74% in 2001 (generation 1), and 19% in 2001 (gen- edge, and this effect was independent of host density. eration 2). The only factor that had a consistent in¯u- However, the mechanisms generating this edge effect June 2003 PARASITOID RESPONSE TO THE MATRIX 1513 differed between cordgrass patches embedded in mud- have a substantial impact on interpatch movement of ¯at and nonhost grass matrices. In mud¯at-embedded this parasitoid. In a nonhost grass matrix, A. columbi patches, similar per capita numbers of hosts parasitized captures were reasonably well described by a simple between the patch edge and interior (Appendix C), cou- diffusion model, a common pattern of redistribution pled with higher A. columbi densities in the interior, for insect species (Turchin 1998). Movement by A. col- likely explained the edge effect (see Fig. 2A). For umbi is predominantly at or below the vegetation can- patches embedded in native grasses or brome, a greater opy; traps positioned above the canopy capture few proportion of hosts parasitized in the interior relative individuals (unpublished data). In contrast, of the few to the edge, was likely brought about by an even dis- female A. columbi that emigrated into a mud¯at matrix, tribution of A. columbi within the patch coupled with captures were distant independent and nondiffusive. a 71% higher per capita attack rate in the patch interior This pattern of captures-with-distance from a source than the edge. At this point, I do not know the cause would be expected if insects displayed nonrandom di- for matrix-dependent edge effects in A. columbi ovi- rectional movements with long step lengths (see Tur- position behavior. One possibility is that factors that chin et al. 1991). Linear movement paths in resource- disrupt oviposition behavior (e.g., predators or wind free or open matrices may be a common pattern among gusts) may be greater at the patch edge than the interior foraging insects (Zalucki and Kitching 1982, Jonsen (e.g., Bowers and Dooley 1993, Burkey 1993). In mud- and Taylor 2000). Under these circumstances, coloni- ¯at edges, those factors may cause A. columbi to dis- zation rates of nearby patches embedded in a mud¯at perse to the patch interior, whereas in nonhost grass matrix likely would be lower than in a nonhost grass edges those factors may serve only to reduce successful matrix where movement is more diffusive, provided ovipositions. Another possibility is interference among that long-range attraction to host sources is similar be- searching parasitoids (Visser et al. 1999, Hassell 2000). tween matrix types. This is precisely the result ob- For all matrix types combined, A. columbi density served in my experiment: relative colonization rates of among interior and edge transects was negatively cor- cordgrass patches located 3 m away from a source patch related with per capita hosts parasitized. Parasitoid in- were more than ®ve times greater in brome and two terference, however, could not have accounted for the times greater in native grasses than in mud¯ats. low per capita attacks 3 m into the mud¯at and native To date, few landscape studies have looked beyond grass matrices where A. columbi densities were also simply quantifying turnover rates among patches em- low (Fig. 2B). bedded in different matrices and have attempted to Edge effects, such as those found with regard to A. quantify the behavioral bases for matrix effects (but columbi densities and parasitism, are evident in a num- see Crist et al. 1992, Ims 1995, Wiens et al. 1995, 1997, ber of other arthropod-predator systems (e.g., Mc- Jonsen and Taylor 2000). The present study is only a Geoch and Gaston 2000, Tscharntke et al. 2002). The ®rst step toward understanding the mechanistic basis potential population consequences of an edge effect can for how the matrix in¯uences A. columbi emigration be manifold (Fagan et al. 1999). In this study system, (e.g., mud¯at edge is less permeable than a brome both the planthopper and A. columbi exhibit a signif- edge), and interpatch movement (e.g., diffusion in non- icant density edge effect only in mud¯at-embedded host grasses, something else entirely in mud¯ats). Be- patches; however, they have opposite responses to a havioral studies at the individual level are needed to mud¯at edge (this study, see Results; K. J. Haynes and further resolve the mechanisms underlying differences J. T. Cronin, unpublished manuscript). Planthoppers in matrix resistance to dispersing A. columbi (see also tend to aggregate at the edge, and A. columbi in the Gustafson and Gardner 1996). interior, of cordgrass patches. By altering the distribution (or ratio) of A. columbi relative to its host, the Anagrus columbi spatial distribution matrix may indirectly in¯uence a number of factors It is clear from the ®ve-generation survey of cord- linked to host±parasitoid stability, including parasitoid grass patches within a 65-ha prairie fragment that the functional responses (Kaiser 1983, Abrams 1993), prey distribution of A. columbi is tightly linked to the dis- outbreaks (Kareiva and Odell 1987), and the distri- tribution of its host, eggs of P. crocea. In all ®ve gen- bution of parasitism risk (Hassell et al. 1991, Pacala erations, almost 60% of the variation in A. columbi and Hassell 1991). In a separate study (Cronin 2003a), abundance (parasitized hosts) was explained by host I found that the distribution of parasitism risk (the co- density. Lei and Hanski (1997) similarly found that ef®cient of variation squared in the probability of par- abundances of the braconid parasitoid Cotesia meli- asitism among cordgrass leaves) was much more het- taearum and its host, the Glanville fritillary Melitaea erogeneous on the patch edge than the interior; in- cinxia, were highly correlated among patches of dry creased heterogeneity in parasitism can be theoretically meadow. stabilizing (Pacala and Hassell 1991; but see Murdoch Features of the landscape were also important in de- and Stewart-Oaten 1989, Gross and Ives 1999). termining the distribution of A. columbi, but their com- In addition to its effect on within-patch distributions bined effects accounted for only ϳ12% of the variation and behavior of A. columbi, the matrix also appears to in abundances among patches. I found that the greater 1514 JAMES T. CRONIN Ecology, Vol. 84, No. 6 the percentage of the nearby matrix that was mud¯at, in a mud¯at matrix and high in a brome matrix; native the more dense the local A. columbi population. Be- grasses tend to have intermediate connectivity. cause A. columbi is less inclined to emigrate from In landscapes dominated by native matrix types (na- patches in mud¯at than in nonhost grasses, it was ex- tive grasses and mud¯at), the spatial structure of P. pected that local populations would build to higher crocea is best described as a mainland±island meta- levels in the former than the latter patches. Of equal population with moderately high connectivity (Cronin importance to the matrix in determining A. columbi 2003b). Because A. columbi appears to track host abun- abundance was patch isolation. In contrast to the ®nd- dances, it likely exhibits the same population structure, ings of many other studies (reviewed in Hanski 1999), although, A. columbi is more dispersal limited than its isolated cordgrass patches tended to harbor greater A. host (Cronin 2003a). For landscapes in which brome columbi densities. This counterintuitive result is un- has displaced native matrix vegetation, the expected derstandable in light of the dispersal and oviposition increase in patch connectivity (see above) is likely to behavior of A. columbi. I (Cronin 2003a) previously cause a shift in the spatial structure of these two species found that the number of A. columbi immigrants de- from a mainland±island metapopulation to a patchy clined with distance from a source patch. A reduction population (Harrison and Taylor 1997, Thomas and Ku- in number of colonists in an isolated patch, however, nin 1999). Increased patch connectivity due to this was compensated for by a substantial increase in per change in matrix composition is predicted to lead to capita oviposition rates (see also Cronin and Strong high spatial synchrony, low host abundances, and low 1999). If isolation further reduced the propensity to interaction persistence at a regional scale (Holyoak and emigrate, a buildup of A. columbi densities that is great- Lawler 1996, Harrison and Taylor 1997, Holyoak 2000, er than in nonisolated patches would be expected. Fi- Nachman 2001). Exotic plant species such as smooth nally, the absence of a patch size effect on A. columbi brome, which are becoming prevalent components of density corroborates previous experimental studies that ecosystems worldwide (Drake et al. 1989, D'Antonio found only weak effects of patch size on A. columbi and Vitousek 1992), may be playing an important role immigration and oviposition behavior (Cronin 2003a). in affecting the spatial structure and dynamics of pop- These studies reveal the importance of experimental ulations of native herbivores and their natural enemies. data in interpreting insect distributional patterns in na- ACKNOWLEDGMENTS ture (see also Lei and Camara 1999, Doak 2000a, b). The completion of this study was made possible by the assistance of the following people: R. Beasler, D. Cronin, F. Host±parasitoid patch connectivity Dillimuth, J. Geber, K. Haynes, T. Hanel, S. Jorde, C. Rhodes, R. Shefchik, M. Szymanski, A. Widdell, and M. Williams. Predator±prey persistence in a spatially subdivided K. Tompkins of the U.S. Fish and Wildlife Service granted habitat can depend critically upon the relative rates of me permission to work at the Kelly's Slough National Wildlife dispersal of both species (e.g., Reeve 1988, Comins et Refuge. I am also grateful to R. Hendersen, H. Johnson, and K. Quanrud for access to their land. K. Haynes, J. Reeve, al. 1992, Holyoak and Lawler 1996). The intervening and two anonymous reviewers provided valuable comments matrix may therefore greatly affect population dynam- on earlier drafts of this manuscript. This work was supported ics by differentially in¯uencing patch connectivity for by the University of North Dakota, ND EPSCoR (EPS- a prey (host) species and its predator (parasitoid). In a 9874802), the UND Alumni Foundation, the City of Grand Forks, Louisiana State University, and the National Science companion study, K. J. Haynes and J. T. Cronin (un- Foundation (DEB-9973789). published manuscript) experimentally quanti®ed the effects of the matrix on P. crocea emigration, immi- LITERATURE CITED gration, and within-patch distribution. In general, P. Abrams, P. A. 1993. Why predation rates should not be pro- crocea and A. columbi transfer processes were affected portional to predator density. Ecology 74:726±733. Bolger, D. T., A. V. Suarez, K. R. Crooks, S. A. Morrison, in qualitatively the same way by matrix heterogeneity. AND T. J. Case. 2000. Arthropods in urban habitat frag- Mark±release±recapture experiments revealed that ments in southern California: area, age, and edge effects. patches embedded in mud¯at have lower planthopper Ecological Applications 10:1230±1248. emigration and immigration rates than comparable Bowers, M. A., and J. L. Dooley, Jr. 1993. Predation hazard and seed removal by small mammals: microhabitat versus patches embedded in brome or native grasses. In the patch scale effects. Oecologia 94:247±254. present study, the scarcity of A. columbi at the patch Brown, J. H., and A. Kodric-Brown. 1977. Turnover rates in edge and in the mud¯at matrix may indicate a low insular biogeography: effect of immigration on extinction. emigration rate, or alternatively, the rapid disappear- Ecology 58:445±449. Burkey, T. V. 1993. Edge effects in seed and egg predation ance of emigrant A. columbi from the vicinity of a at two Neotropical rainforest sites. Biological Conservation patch. In contrast, higher A. columbi densities in the 66:139±143. brome than mud¯at matrix may have been due to high Cantrell, R. S., and C. Cosner. 1999. Diffusion models for emigration rates or a propensity for emigrants to remain population dynamics incorporating individual behavior at boundaries: applications to refuge design. Theoretical Pop- near their patch of origin. At the spatial scale of several ulation Biology 55:189±207. meters from a source patch, these data suggest that Cappuccino, N., and M. A. Martin. 1997. The birch tube- patch connectivity for P. crocea and A. columbi is low maker Acrobasis betulella in a fragmented habitat: the im- June 2003 PARASITOID RESPONSE TO THE MATRIX 1515

portance of patch isolation and edges. Oecologia 110:69± Ims, R. A. 1995. Movement patterns related to spatial struc- 76. tures. Pages 85±109 in L. Hansson, L. Fahrig, and G. Mer- Cappuccino, N., and R. B. Root. 1992. The signi®cance of riam, editors. Mosaic landscapes and ecological processes. host patch edges to the colonization and development of Chapman and Hall, London, UK. Corythucha marmorata (Hemiptera: Tingidae). Ecological Jonsen, I. D., R. S. Bourchier, and J. Roland. 2000. The Entomology 17:109±113. in¯uence of matrix habitat on aphthona ¯ea beetle immi- Comins, H. N., M. P. Hassell, and R. M. May. 1992. The gration to leafy spurge patches. Oecologia 127:287±294. spatial dynamics of host±parasitoid systems. Journal of An- Jonsen, I. D., and P. D. Taylor. 2000. Fine-scale movement imal Ecology 61:735±748. behaviors of calopterygid damsel¯ies are in¯uenced by Crist, T. O., D. S. Guertin, J. A. Weins, and B. T. Milne. 1992. landscape structure: an experimental manipulation. Oikos Animal movement in heterogeneous landscapes: an exper- 88:553±562. iment with Eleodes beetles in shortgrass prairie. Functional Kaiser, H. 1983. Small scale spatial heterogeneity in¯uences Ecology 6:536±544. predation success in an unexpected way: model experi- Cronin, J. T. 2003a. Patch structure, oviposition behavior, ments on the functional response of predatory mites (Ac- and the distribution of parasitism risk. Ecological Mono- arina). Oecologia 56:249±256. graphs 73:283±300. Kareiva, P. 1985. Finding and losing host plants by Phyllo- Cronin, J. T. 2003b. Movement and spatial population struc- treta: patch size and surrounding habitat. Ecology 66: ture of a prairie planthopper. Ecology 84:1179±1188. 1809±1816. Cronin, J. T., and D. R. Strong. 1999. Dispersal-dependent Kareiva, P., and G. Odell. 1987. Swarms of predators exhibit oviposition and the aggregation of parasitism. American ``preytaxis'' if individual predators use area-restricted Naturalist 154:23±36. search. American Naturalist 130:233±270. Krombein, K. V., P. D. Hurd, Jr., D. R. Smith, and B. D. D'Antonio, C. M., and P. M. Vitousek. 1992. Biological in- Burks. 1979. Catalog of Hymenoptera in America north of vasions by exotic grasses, the grass/®re cycle and global Mexico. Smithsonian Institution Press, Washington, D. C., change. Annual Review of Ecology and Systematics 23: USA. 63±87. Lei, G. C., and M. D. Camara. 1999. Behaviour of a specialist Doak, P. 2000a. The effects of plant dispersion and prey parasitoid, Cotesia melitaearum: from individual behaviour density on parasitism rates in a naturally patchy habitat. to metapopulation processes. Ecological Entomology 24: Oecologia 122:556±567. 59±72. Doak, P. 2000b. Population consequences of restricted dis- Lei, G. C., and I. Hanski. 1997. Metapopulation structure of persal for an insect herbivore in a subdivided habitat. Ecol- Cotesia melitaearum, a specialist parasitoid of the butter¯y ogy 81:1828±1841. Melitaea cinxia. Oikos 78:91±100. Drake, J. A., F. DiCastri, R. H. Groves, F. J. Kruger, H. A. Lovejoy, T. E., J. M. Rankin, R. O. Bierregaard, Jr., K. S. Mooney, M. Rejmanek, and M. H. Williamson. 1989. Bi- Brown, L. H. Emmons, and M. E. van der Voort. 1984. ological invasions: a global perspective. Wiley, New York, Ecosystem decay of Amazon forest fragments. Pages 295± New York, USA. 325 in M. H. Nitecki, editor. Extinctions. University of Fagan, W. E., R. S. Cantrell, and C. Cosner. 1999. How Chicago Press, Chicago, Illinois, USA. habitat edges change species interactions. American Nat- McGeoch, M. A., and K. J. Gaston. 2000. Edge effects on uralist 153:165±182. the prevalence and mortality factors of Phytomyza ilicis Gross, K., and A. R. Ives. 1999. Inferring host±parasitoid (Diptera, Agromyzidae) in a suburban woodland. Ecology stability from patterns of parasitism among patches. Amer- Letters 3:23±29. ican Naturalist 154:489±496. Moilanen, A., and I. Hanski. 1998. Metapopulation dynam- Gustafson, E. J., and R. H. Gardner. 1996. The effect of ics: effects of habitat quality and landscape structure. Ecol- landscape heterogeneity on the probability of patch colo- ogy 79:2503±2515. nization. Ecology 77:94±107. Murdoch, W. W., and A. Stewart-Oaten. 1989. Aggregation Hanski, I. 1999. Metapopulation ecology. Oxford University by parasitoids and predators: effects on equilibrium and Press, New York, New York, USA. stability. American Naturalist 134:288±310. Harrison, S., and A. D. Taylor. 1997. Empirical evidence for Nachman, G. 2001. Predator±prey interactions in a non- metapopulation dynamics. Pages 27±42 in I. Hanski and equilibrium context: the metapopulation approach to mod- M. E. Gilpin, editors. Metapopulation biology: ecology, eling ``hide-and-seek'' dynamics in a spatially explicit tri- genetics, and evolution. Academic Press, San Diego, Cal- trophic system. Oikos 94:72±88. ifornia, USA. Okubo, A. 1980. Diffusion and ecological problems: math- Hassell, M. P. 2000. The spatial and temporal dynamics of ematical models. Springer-Verlag, Heidelberg, Germany. host±parasitoid interactions. Oxford University Press, New Pacala, S. W., and M. P. Hassell. 1991. The persistence of York, New York, USA. host±parasitoid associations in patchy environments. II. Hassell, M. P., R. M. May, S. W. Pacala, and P. L. Chesson. Evaluation of ®eld data. American Naturalist 138:584±605. Reeve, J. D. 1988. Environmental variability, migration, and 1991. The persistence of host±parasitoid associations in persistence in host±parasitoid systems. American Naturalist patchy environments. I. A general criterion. American Nat- 132:810±836. uralist 138:568±583. Ricketts, T. H. 2001. The matrix matters: effective isolation Holder, M. W., and S. W. Wilson. 1992. Life history and in fragmented landscapes. American Naturalist 158:87±99. descriptions of the immature stages of the planthopper Pro- Roland, J., and W. J. Kaupp. 1995. Reduced transmission of kelisia crocea (Van Duzee) (Homoptera: Delphacidae). forest tent caterpillar (Lepidoptera: Lasiocampidae) nuclear Journal of the New York Entomological Society 100:491± polyhedrosis virus at the forest edge. Environmental En- 497. tomology 24:1175±1178. Holyoak, M. 2000. Habitat patch arrangement and metapop- Roland, J., N. Keyghobadi, and S. Fownes. 2000. Alpine ulation persistence of predators and prey. American Nat- Parnassius butter¯y dispersal: effects of landscape and uralist 156:378±389. population size. Ecology 81:1642±1653. Holyoak, M., and S. P. Lawler. 1996. The role of dispersal Simms, E. L., and D. S. Burdick. 1988. Pro®le analysis of in predator±prey metapopulation dynamics. Journal of An- variance as a tool for analyzing correlated responses in imal Ecology 65:640±652. experimental ecology. Biometrical Journal 30:229±242. 1516 JAMES T. CRONIN Ecology, Vol. 84, No. 6

Sokal, R. R., and F. J. Rohlf. 1995. Biometry. Third edition. Turchin, P., and W. T. Thoeny. 1993. Quantifying dispersal W. H. Freeman, New York, New York, USA. of southern pine beetles with mark±recapture experiments Stamps, J. A., M. Buechner, and V. V. Krishnan. 1987. The and a diffusion model. Ecological Applications 3:187±198. effects of edge permeability and habitat geometry on em- Visser, M. E., T. H. Jones, and G. Driessen. 1999. Interference igration from patches of habitat. American Naturalist 129: among insect parasitoids: a multi-patch experiment. Journal 533±552. of Animal Ecology 68:108±120. Taylor, P. D., L. Fahrig, K. Henein, and G. Merriam. 1993. Wiens, J. A. 1997. Metapopulation dynamics and landscape Connectivity is a vital element of landscape structure. Oi- ecology. Pages 43±67 in I. A. Hanski and M. E. Gilpin, kos 68:571±573. editors. Metapopulation biology: ecology, genetics, and Thomas, C. D., and W. E. Kunin. 1999. The spatial structure evolution. Academic Press, San Diego, California, USA. of populations. Journal of Animal Ecology 68:647±657. Wiens, J. A., T. O. Crist, and B. T. Milne. 1993. On quan- Tilman, D., and P. Kareiva. 1997. Spatial ecology: the role tifying insect movement. Environmental Entomology 22: of space in population dynamics and interspeci®c inter- 709±715. actions. Princeton University Press, Princeton, New Jersey, Wiens, J. A., T. O. Crist, K. A. With, and B. T. Milne. 1995. USA. Fractal patterns of insect movement in microlandscape mo- Tscharntke, T., I. Steffan-Dewenter, A. Kruess, and C. Thies. saics. Ecology 76:663±666. 2002. Contribution of small habitat fragments to conser- Wiens, J. A., R. L. Schooley, and R. D. Weeks. 1997. Patchy vation of insect communities of grassland±cropland land- landscapes and animal movements: do beetles percolate? scapes. Ecological Applications 12:354±363. Oikos 78:257±264. Turchin, P. 1998. Quantitative analysis of movement: mea- Wilson, S. D., and J. W. Belcher. 1989. Plant and bird com- suring and modeling population redistribution in animals munities of native prairie and introduced Eurasian vege- and plants. Sinauer Associates, Sunderland, Massachus- tation in Manitoba, Canada. Conservation Biology 3:39± setts, USA. 44. Turchin, P., F. J. Odendaal, and M. D. Rausher. 1991. Quan- Zalucki, M. P., and R. L. Kitching. 1982. The analysis and tifying insect movement in the ®eld. Environmental En- description of movement in adult Danaus plexippus L. tomology 20:955±963. (Lepidoptera: Danainae). Behaviour 80:174±198.

APPENDIX A Details on creation of host pots and trap pots for use in estimating A. columbi parasitism and density, respectively, are presented in ESA's Electronic Data Archive: Ecological Archives E084-039-A1.

APPENDIX B Census procedure and analyses of A. columbi distributions among cordgrass patches are provided in ESA's Electronic Data Archive: Ecological Archives E084-039-A2.

APPENDIX C A table of statistical results for contrasts between the patch interior and edge, and interior and matrix for three A. columbi response variables (density, proportion parasitized, and per capita hosts parasitized) is provided in ESA's Electronic Data Archive: Ecological Archives E084-039-A3.